RSD Fault Block Model for Highly Efficient Fault-Tolerant Manhattan Routing Algorithms in 2D Mesh

2016 ◽  
Vol 59 (10) ◽  
pp. 1511-1526 ◽  
Author(s):  
Hongzhi Zhao ◽  
Yuan Xue
2017 ◽  
Vol 27 (04) ◽  
pp. 1850054 ◽  
Author(s):  
Hongzhi Zhao ◽  
Qiang Wang ◽  
Ke Xiong ◽  
Songwen Pei

Fault-tolerant Manhattan routing algorithms aim at finding a Manhattan path between the source and destination nodes and route around all faulty nodes. However, besides faulty nodes, some nonfaulty nodes that are helpless to make up a fault-tolerant Manhattan path should also be routed around. How to label such nonfaulty nodes efficiently is a major challenge. We propose a path-counter method. It can label such nodes with low time-complexity by counting every node’s fault-tolerant Manhattan paths to the source or destination node. During the path-counting procedure, no available nodes will be sacrificed under arbitrary fault distribution. Compared with fault-block model based work, our proposed method is independent of fault distribution, so its computational complexity is very low.


2019 ◽  
Vol 28 (04) ◽  
pp. 1950055 ◽  
Author(s):  
Munshi Mostafijur Rahaman ◽  
Prasun Ghosal ◽  
Tuhin Subhra Das

Reliability of a Network-on-Chip (NoC) relies vastly upon the efficiency of handling faults. Faults those lead to trouble during on-chip communication process are basically of two types namely soft and hard. Here, hard faults are considered. Hard faults may be caused due to failure of links, routers, or other processing units. These are mainly dealt with fault-tolerant routing algorithms or by employing redundant hardware. Multiple faulty nodes are being avoided by acquiring region-based approaches. Most of the fault-tolerant routing techniques are designed on homogeneous faulty regions where some active nodes also act as deactivated nodes to build the region homogeneous. On the other hand, adaptive routing on nonhomogeneous faulty regions increases load on its boundary and most of them does not assure deadlock freeness. This paper proposes a deadlock-free adaptive fault-tolerant NoC routing named F-Route-NoC-Mesh (FRNM) ignoring any virtual channel on orthogonal convex faulty regions. Contributions of this work focus on balancing network traffic by assuming a virtual faulty block boundary and routing packets through this virtual boundary. Destination does not exist within that virtual faulty block regions to reduce load on the boundary of orthogonal faulty regions. Thus, this work is aimed at acquiring proper incorporation of procedures being able to reach fault-tolerant degree, routing efficiency and performance enhancement. Using the proposed algorithm (FRNM), a fault block model-based approach is developed. Significant improvements of average latency (43.37% to 60.44%), average throughput (4.18% to 90.81%) and power consumption (5.93% to 33.28%) are achieved over the state-of-the-art by using a cycle accurate simulator.


2007 ◽  
Vol 53 (12) ◽  
pp. 902-912 ◽  
Author(s):  
Huaxi Gu ◽  
Jie Zhang ◽  
Kun Wang ◽  
Zengji Liu ◽  
Guochang Kang

Sign in / Sign up

Export Citation Format

Share Document